Calcium imaging in freely moving mice during electrical stimulation of deep brain structures (2024)

In total 30 adult C57BL/6 J male mice approximately 25–30 g (Jackson Laboratories, Bar Harbor, ME) were used to test each component of the experimental protocol. These animals were used for verifying 6-OHDA lesioning and behavior (n = 15, only shown in supplementary data available online at stacks.iop.org/JNE/18/026008/mmedia), and chronic recordings in healthy mice and parkinsonian mice (n = 5 and n = 10, respectively). Of the ten parkinsonian mice, four were removed from analysis due to surgical complications and implant/expression issues while the others were used for data collection while under anesthesia (n = 6). Two these six animals in which anesthetized recordings were performed were removed from the analyses presented here, as post-mortem histology indicated their electrodes were not properly placed within STN. Awake calcium imaging studies were also performed in two of these animals. Exclusion criteria and experimental outcomes are described in table 1.

Animals were group-housed and maintained under a standard light/dark cycle with ad libitum access to food and water. All experimental procedures were approved by the either the Mayo Clinic Institutional Animal Care and Use Committee (IACUC) or the IACUC at LifeSource Biomedical Services, NASA Ames Research Center, California and were carried out in accordance with the U.S. National Research Council Guide for the Care and Use of Laboratory Animals.

Mice were anesthetized in an induction chamber with 5% isoflurane (inhalation) and then maintained with a continuous flow of 1%–2% isoflurane-oxygen mixture during stereotactic surgery (Kopf Systems, Tujunga, CA). A midline incision was performed to expose the skull and burr holes were made above the dorsal striatum (ML = 1.65, AP = 1.0 from bregma) and substantia nigra pars compacta (SNc) (ML = 1.25, AP = –3.5 from bregma). A Hamilton syringe (Hamilton Company, Reno, NV) controlled by a stereotactically mounted microdrive (WorldPrecision Instruments) was lowered into right dorsal striatum (AP = +1.0 mm, ML = +1.65 mm, DV = −2.7 mm from dura) at an approximate rate of 10 µm s⁻¹ and used to inject 500 nL of AAV9.CAG.GCaMP6m.WPRE.bGHpA (titer = 5.5 × 10¹² GC mL⁻¹; SignaGen Laboratories, Rockville, MD) at a rate of 100 nl min⁻¹ (figure 2(B)). Brain tissue was given one minute to settle prior to the unilateral AAV injection. The virus was allowed to diffuse for an additional five minutes prior to removing the needle at approximately 10 µm s⁻¹.

Immediately following AAV injection, a separate Hamilton syringe was used to inject 6-OHDA (4 µg µl⁻¹) into the SNc at three depths: DV −4.2, −4, and −3.8 from dura (figure 2(B)). Injections at each location were 650 nL, and delivered at a rate of 100 nl min⁻¹. At each of the three depths, the injection was preceded with a wait period of one minute to allow brain tissue to stabilize, and followed by a wait period of five minutes for diffusion [43]. Insertion, removal and movements of the injection needle were carried out at approximately 10 µm s⁻¹. Desipramine HCl was injected (25 mg kg⁻¹, i.p.) approximately 30 min before the first 6-OHDA injection to prevent uptake of 6-OHDA into monoamine neurons [47]. The burr holes were sealed with Kwik-Sil (World Precision Instruments, Sarasota, FL) before suturing the skin incision. Ketofen (2.5 mg kg⁻¹, i.p.) and Carprofen (2.5 mg kg⁻¹, i.p.) were administered to reduce inflammation and provide analgesia. Each animal was allowed to recover individually in a heated home cage until fully mobile and were then returned to social housing.

The full experimental paradigm, consisting of 6-OHDA lesioning, AAV9 injection, GRIN lens implantation and STN stimulating electrode implantation, was performed on ten mice. Two mice were lost due to surgical complications, one due to GRIN lens damage that prevented imaging, and one animal that had insufficient GCaMP expression for imaging. An additional two mice were excluded from all analyses due to mistargeting of the stimulating electrode outside of the STN as assessed by post-mortem histology. These exclusion criteria are described in table 1. Expression of GCaMP6m calcium indicator, induced by AAV9.CAG.GCaMP6m.WPRE.bGHpA injection, was confirmed via histological analysis that showed green fluorescent cell bodies in the dorsal striatum (figure 3(B)). Most GCaMP expressing cells did not show any nuclear expression, which is an important indicator of cell health (figure S3 (available online at stacks.iop.org/JNE/18/026008/mmedia)) [48]. Dopaminergic lesioning via a unilateral 6-OHDA lesion, performed in the ipsilateral hemisphere, was confirmed by immunohistochemical labeling of TH followed by quantification of TH positive cells and mean pixel intensity (figure 3(A)). A significant decrease in dorsal striatal mean pixel intensity (figure 3(D), p < 0.05; Welch's unequal variance t-test), and SNc TH positive cells (figure 3(E), p < 0.05; Welch's unequal variance t-test) between the lesioned and non-lesioned hemispheres was observed.

Initial validation of calcium imaging within the striatum was performed in healthy mice without stimulation electrode implants (figure 4). A consistent number of cells, as identified by CNMF-E, were obtained in six recording sessions performed over 33 d across five animals (figure 4(B)). This indicates that recordings from the same brain area can be performed over time, however slight differences in the field of view were observed between recording sessions (figure 4(C)). Additionally, damage to the microscope during a recording session and the subsequent repair led to a slight change in the focal plane, including a rotation of 90°, between sessions (figure 4(C)).

Initial validation of the ability to record STN stimulation evoked responses in striatum prior to awake behaving experiments was performed in six anesthetized animals. Two animals did not exhibit stimulation evoked response within striatum; it was later verified via post-mortem histology that neither of the electrodes of the twisted bipolar pair were located in STN, and these animals were removed from the analysis. Stimulation evoked responses were observed in four anesthetized animals that were imaged during 10 s epochs of 30, 80 and 130 Hz stimulation. These epochs were separated by 1-min intervals to eliminate the residual effects from previous stimulations. This sequence was repeated three times in each animal. The striatum was quiescent while stimulation was turned off (figure 5(A)) under anesthesia, and only became active when STN stimulation was applied (figures 5(B)–(D)). STN stimulation activated a subset of in-focus cells within the striatum as well as caused out-of-focus changes in background fluorescence (figures 5(B)–(D)), which may originate from stimulation-evoked broad changes in cell activation. These stimulation-dependent out-of-focus signals are exemplified by changes in mean fluorescence within the field-of-view shown in figure 5(E). These out-of-focus changes and synchronous activation of multiple neurons confounded the analysis of calcium signals via standard data analysis techniques such as PCA/ICA, which identifies cells using differences in temporal characteristics (figure S5(B)). To quickly assess the performance of common algorithms in isolating activity of individual cells when in the presence of significant out of plane calcium activity induced by stimulation, simulated changes in background fluorescence based on a model of out-of-plane activation were added to an experimental data set (figure S1). The capability of common data analysis methods and spatial filtering techniques to identify spatially compact neurons and remove the simulated contamination were then qualitatively compared (figure S5).

ROI analysis, PCA/ICA with or without spatiotemporal de-mixing, and CNMF-E [55, 56] were evaluated for their ability to extract well-defined cell bodies in the presence of real and simulated background fluorescence. Although ROI analysis allows for selection of well-defined cell bodies, it is a subjective technique and cannot separate cell activity from low spatial frequency background contamination. As a result, calcium traces extracted during electrical stimulation of the STN were nearly identical for all cells in the field of view (figure S5(A)). This suggests that the traces were primarily the results of background contamination and not from individual neural activity. Additionally, ROI analysis was not able to remove simulated background contamination (figure S5(A)). These results suggest that ROI analysis is not well suited for neuron identification from recordings obtained during electrical stimulation. Similar to ROI analysis, PCA/ICA [64] failed at uniquely identifying cells that exhibited synchronous firing during stimulation and was unable to completely remove artificial background contamination (figure S5(B)). Spatiotemporal de-mixing, which can be used to improve the identification of cells with PCA/ICA [64], did not result in significant improvements (figure S5(C)). Both ROI analysis and PCA/ICA were applied to the same data prefiltered using a spatial bandpass filter with cutoff frequencies chosen based on the size of cell bodies observed in the ΔF/F data (5 and 31 pixels). In both cases spatial filtering improved the rejection of artificial background contamination but did not completely eliminate the background contamination (figures S5(D)–(F)). In contrast, CNMF-E was able to identify individual cell bodies despite synchronous neural activity (figure S5(G)), does so in an objective manner, and is much less susceptible to low spatial frequency background noise. For this reason, we used CNMF-E for all subsequent data analysis presented in this manuscript.

A preliminary characterization of frequency dependent changes in STN stimulation evoked calcium responses within striatum was performed on recordings from four anesthetized mice with stimulation electrodes located in the STN that were described in the previous section. Although the striatum was observed to be quiescent under anesthesia (figure 5(A)), clear calcium activity suggestive of neuronal activation occurred with onset of STN stimulation at 30, 80, and 130 Hz (figures 5(B)–(D), Video S1). Additionally, the recruitment occurred in a frequency-dependent manner (figures 6(A)–(C)) with 80 and 130 Hz stimulation exhibiting a greater number of activated cells than stimulation at 30 Hz (figure 6(D), p < 0.05; Kruskal–Wallis one-way analysis of variance, Dunn's multiple comparison test). Despite the overall frequency dependence of neural activation, there was heterogeneity in the response of individual neurons to different stimulations as shown by figure 6(E). For instance, some cells only responded to stimulation at 80 Hz, but not 130 Hz (i.e. Figure 6(E), cell 2) or even showed heterogeneous responses to repeated stimulation at the same parameters (i.e. Figure 6(E), cell 4). The stimulation evoked behavior of specific cells did not necessarily depend on relative location of the cell in the field of view, with co-localized neurons showing different responses to similar parameters (figure 6(E)).

Two of the four mice in which anesthetized stimulation evoked recordings were performed also underwent awake-behaving imaging sessions in an open-field during stimulation (200 µs pulsewidth, 30 µA amplitude and 200 µs pulsewidth, 80 µA amplitude). In contrast to anesthetized recordings in the same animal, the striatal neural population was found to be highly active during open field behavior with increased activity during animal movement (figure 7(A), Video S2). This can be seen in the relationship between animal velocity (figure 7(A) top row) and cell activity shown on the raster plots (figure 7(A) second row) and calcium traces (figure 7(A) third row). The raster plot shown in the second row of figure 7(A) shows the calcium event timing for all cells in a representative data set and the traces show) and the raster of calcium traces on the third row of figure 7(A) shows the calcium activity for the 40 highest SNR cells in the data set. Significant differences in the relationship between the rate of calcium events and animal velocity were observed at both 30 Hz figures 7(B), (C); p < 0.0001 Kruskal–Wallis one-way analysis of variance, Dunn's multiple comparison test) and 130 Hz (figures 7(B), (C); p < 0.0001 Kruskal–Wallis one-way analysis of variance, Dunn's multiple comparison test).

The fundamental goal of the experiments outlined in this methods paper was to demonstrate technical feasibility of the multicomponent surgical procedure, as well as demonstrate that the stimulation evoked responses in a commonly explored downstream structure (striatum) of a traditional DBS target (STN) varied as a function of behavioral state (anesthetized, awake sedentary, awake moving). These data were intended to underscore that the combined experimental preparation can yield unique information that may be useful in the study of DBS mechanisms in the future, but were not intended at this time to make any definitive conclusions about the therapeutic mechanisms of STN DBS to treat parkinsonian symptoms.

As a first exploration of the capabilities of single photon calcium imaging to assess the effects of DBS, we looked at the frequency dependence of evoked neural activity in anesthetized mice. Other studies have shown that electrical stimulation excites neural elements in a frequency dependent manner [65] and that therapeutic responses to neuromodulation are highly dependent on stimulation frequency [66]. For example, therapeutic efficacy for STN DBS in human patients is typically achieved with high frequency stimulation in the range of 120–180 Hz [67]. The population of neurons activated by stimulation during anesthesia was a small subset of the neurons identified during unstimulated awake behaving recordings (figures 6(A)–(C)). This sparse activation is consistent with other work using optical techniques showing sparse neural activation around a stimulation electrode [65, 68]. An increase in the number of cells activated at 80 and 130 Hz stimulation was observed compared to stimulation at 30 Hz (figure 6(D)). In anesthetized optical recordings, heterogeneous changes in neural activation within striatum were observed, with different cells that were near each other in the microscope field of view responding to stimulation at different frequencies (figure 6(E)). This heterogeneity may be due to differences in response of specific cell types. The heterogeneity could also be the result of subtle changes in stimulation evoked local neuronal/non-neuronal activity as a function of distance from the bipolar electrode, leading to varied responses at a distal point in the connected circuit.

To illustrate the technical feasibility of performing optical recordings in striatum while stimulating STN in an awake behaving pathological model, we performed a series of recordings with 10-min periods of 30 Hz and 130 Hz electrical stimulation interleaved by 10-min periods without stimulation. In behaving mice, the striatum was found to be much more active (figure 7(A)) than under anesthesia (figure 5(A)). Additionally, neuronal activity within the striatum was found to increase with movement of the mouse, as assessed by animal velocity (figures 7(A)–(B)). In contrast to STN stimulation in the anesthetized state which increased observed activity in striatum (figure 6), in the two animals in which anesthetized and awake recordings were performed STN stimulation during minimally constrained behavior decreased the activity in striatum (figure 7). Given the low number of animals we cannot draw any definitive conclusions about DBS mechanism across an entire population; however, these data suggest that the downstream effects of DBS may also depend on behavioral inputs to the same circuitry, thus underscoring the need for recording technology compatible with unrestricted behavioral experiments during DBS. Additionally, recording from genetically defined sub-populations of cells using calcium imaging would enable studies of neuromodulation effects in the context of specific neuronal cell types, such as D1 and D2 medium spiny neurons that form the primary outputs of the striatum [43].

The complex nature of these proof of concept experiments led to successful stimulation and imaging in 40% of the animals, with primary failure points being anesthesia related complications during surgery (20%), lens damage during group housing of the animals (10%), insufficient GCaMP expression (10%), and failure to place the STN stimulation electrode within the target structure (20%). These failure points can be improved with practice and some simple modifications to the experimental procedures. However, with multiple cumulative points of failure across different components of this procedure, it is particularly important to minimize failures in order to obtain a practical level of experimental yield. To accomplish this, we recommend optimizing each component of the surgical procedure in isolation to minimize surgical time and complications prior to attempting the components in combination. By optimizing each component in parallel, we were able to complete the first cohort of animals with the entire experimental paradigm approximately two months after beginning this project.

The DBS implant procedure has a high rate of failure due to the difficulty targeting small brain nuclei in the mouse model. For this reason, it is important to practice and optimize the targeting of the DBS implant before combining it with the other components of the overall procedure. Since, validation of electrode location cannot be obtained until histological verification is performed, a large amount of time could be spent performing experiments on animals that will be excluded from the analysis due to mistargeting of the electrode. One step that we have since taken to improve targeting is switching to the use of custom electrodes made from platinum/iridium wire (A-M Systems) instead of the platinum electrodes from PlasticsOne used in these studies. At the electrode sizes used in these experiments (75 µm diameter wire), platinum/iridium electrodes are much stiffer than pure platinum electrodes; in our experience they were less likely to deviate from the target trajectory during insertion. Another component that should be independently validated before attempting implantation of the GRIN lens and subsequent imaging is that GCaMP expression in the target location is sufficient and does not show nuclear expression that is indicative of toxicity [48]. Appropriate GCaMP expression is a critical aspect of successful imaging that can be easily checked before imaging in genetic lines or with AAV induced models. If using an AAV, expression can be adjusted by adjusting the titer of the virus, which may be helpful for optimizing for a specific target location.

An avoidable point of failure in our experiments that can be addressed during the surgical procedures was protection of the GRIN lens from damage during the time between implantation of the lens and baseplating the animal. During the experiments presented in this manuscript, we protected the GRIN lens with a protective plastic cap made from the end of a microcentrifuge tube and attached with Kwik-Cast adhesive. We have since switched to using dental cement to hold the protective cap to the animals headcap for a more secure connection. Whatever method is selected, it should be optimized prior to attempting the complete experimental paradigm to prevent animal losses at this stage.

The implementation of microendoscope optical recordings and simultaneous stimulation through implanted electrodes needs to be carefully planned in the context of specific behavioral experiments, as the connection to instrumentation can both constrain behavior as well as place strain on the connectors and the headcap itself. In our experiments, for instance, individual recording sessions were limited to 10 min in 6-OHDA lesioned mice in order to prevent tangling of the microscope and stimulation wires due to biased rotational behavior of the lesioned animals. In the future this recording time will be extended with the addition of a commutator [69–71]. Appropriate control of the wires and use of a commutator to enable unrestricted behavior of the animal will be critical to rigorous studies of behavioral changes during the application of DBS, where additional wires needed for electrical stimulation can impose additional constraints on behavior of the mouse over the miniature microscope alone.

Other studies using single and two photon microscopy in behaving animals have shown stable longitudinal recordings in excess of one month [43–46]. These data suggest that longitudinal recording from the same cell population over weeks of neuromodulation therapy may be possible. Neuromodulation therapies for some neurologic and psychiatric conditions, such as depression and epilepsy can take months to show therapeutic effect [11, 34–42]. Thus, the capability to longitudinally record from the same set of cells via calcium imaging in a deep brain regions over weeks to months is important to enable the in vivo study of plasticity in neuromodulation therapies that can be difficult via electrophysiology or functional imaging methods. In contrast to electrophysiological recordings where highly similar waveforms measured across time from different neurons in the vicinity of the recording electrode can make it difficult to guarantee the tracking of a single-unit from recording obtained on different days [24, 25], the ability to directly visualize neuron somas in optical recordings and use their relative locations as landmarks may make it easier to confidently track the same neuron over time [72]. In figure 4(B), we demonstrated the ability to record cells during six recording sessions over the course of a month. However, slight differences in the field of view can be seen between recording sessions as well as a rotation of the field of view caused by the repair of the microscope being used in the experiments between days 0 and 12. We attempted to perform tracking of individual cells between recording session in the data from these experiments using the cell tracking algorithm presented in Sheintuch et al (2018). However, this was not successful due to inconsistent placement of the microscope leading to changes in the field of view between sessions. For this reason, although it is likely possible to align cells over longer periods of time up to weeks or months [43–46] in animals undergoing neuromodulation therapies, it is important to note that these analysis require careful placement of the microscope to obtain the same field of view between recording sessions. In practice this proved to be difficult, however the process may be improved by new iterations of microscope technology that include electronic focusing capabilities [73].

Unlike two-photon microscopy, which is robust to changes in activity outside of the focal plane, single photon microscopy can be easily contaminated by out-of-focus changes in neural activity. Although these changes in activity may not significantly impact the analysis of neural activity during some behavioral paradigms, they can have a significant effect when electrical stimulation is applied. Therefore, attention should be paid to application of existing analysis techniques to the study of neuromodulation paradigms, which may introduce non-physiological changes in neural activity. Initially, we applied PCA/ICA, a commonly used analysis technique for calcium imaging data, to identify cell activity in recordings obtained in behaving animals. Using PCA/ICA we observed artifactual high correlation between calcium traces recorded from nearby cells and were not able to distinguish overlapping cells bodies (figure S4), a limitation that have been observed by other recent studies [55, 74]. This led us to perform a qualitative analysis of different algorithms to remove artificially added stimulation induced background fluorescence changes from data recorded in anesthetized animals (figure S5). Our data suggests that ROI analysis and PCA/ICA, which are commonly applied algorithms for identifying neural signals in calcium imaging recordings, were not suitable for isolating calcium traces from in focus neurons from the artificially added out-of-focus contamination. Compared to ROI analysis and PCA/ICA, even with bandpass filtering of the data prior to analysis, CNMF-E was better able to isolate calcium traces from in focus neurons. This suggests that CNMF-E is better suited analysis of data obtained during the application of neuromodulation therapies. These results are in agreement with other comparisons of CNMF-E to PCA/ICA in other contexts that have been recently published [55, 74].

Single photon microscopy has been important for obtaining results in deep brain regions by imaging through a GRIN lens in freely behaving animals [45, 48], however two photon-microscopy can also been performed through a GRIN lens [43, 75]. The minimal out-of-focus fluorescence of two photon microscopy could avoid many of the analysis issues described in this section and has other benefits such as imaging multiple indicators [43], which has not yet become common for single-photon head-mounted systems [76]. The disadvantage of using two-photon microscopy is that less neurons can be simultaneously imaged, and it is generally required to head-fix animals, which can affect the outcomes of behavioral experiments [77].

In order to perform a mechanistic study to isolate the mechanisms of STN DBS therapy with appropriate scientific rigor in the mouse model, there are a number of potential confounds that must be addressed, including: (1) establishing the therapeutic efficacy of DBS to alleviate pathological symptoms in the mouse model without the complication of an invasive indwelling GRIN lens, (2) determining the impact of the indwelling GRIN lens on normal and pathological behavior, and (3) assessing the impact of the indwelling GRIN lens on the therapeutic function of DBS.

Single-photon calcium imaging via a head-mounted microscope is primarily performed in mouse models due to the availability of genetic lines and more extensive experience with induction of expression with viral vectors that can be leveraged to express calcium indicators in specific cell types. In contrast, historic studies of neuromodulation therapies usually utilize rats or large animal models in order to increase space for the implantation of electrodes, improve targeting of small neural structures, and reduce the difference in sizes of electrodes and brain nuclei compared to the human therapies. Although many pathological models for diseases treated by DBS have been well established in mouse models, the electrode configuration and parameters that best mimic clinical DBS in mouse animal models are not well understood [49, 78]. Therefore, the therapeutic efficacy of DBS applied to clinically relevant targets in these models needs to be validated before rigorous studies into the mechanisms of DBS can be performed. We have begun this work for STN-DBS in a 6-OHDA mouse model of Parkinson's disease, with preliminary data provided in the Supplemental Methods as a frame of reference (figure S2).

Another important consideration that needs to be further validated is the effect of tissue aspiration and GRIN lens implantation on the mechanisms underlying the effects of neuromodulation therapies [79]. For example, this procedure disrupts tissue within the motor cortex, striatum, and other structures which may be critical to the therapeutic effects of DBS. There is growing evidence that the trauma due to the implantation of electrodes in or on the brain, and the trauma due to the associated surgical procedure, can induce deficits in motor function and other outcomes. These deficits may only be apparent when more sophisticated measures of motor function are used [79]. Similarly, there is evidence to suggest the presence of a comparatively small chronic indwelling electrode may impact ion channel expression in nearby tissue out as far as 200 microns [80]. For this reason, it is paramount to understand and anticipate potential confounds created by the GRIN lens implant on both normal and pathological behaviors, as well as response to therapeutic interventions. To some extent the lens implant access route can be optimized to minimize influence on pathways putatively associated with the behavior in question, pathological influence on that behavior, and circuitry hypothesized to be engaged in DBS therapeutic effect. DBS therapies are increasingly thought to act through circuit mechanisms that rely on activation of tracts of passage near the electrode in addition to local effects. For instance, one of the current leading theories for the mechanism of STN DBS to treat PD is that DBS disrupts pathological synchrony in the basal ganglia-thalamo-cortical motor circuit through antidromic activation of motor cortex [81, 82]. In general, disruption of these pathways due to damage caused by aspiration and insertion of the GRIN lens has the potential to disrupt or modify the pathological behaviors being studied or the therapeutic mechanism of DBS.

To address these issues for the 6-OHDA mouse model of Parkinson's Disease cohorts directly comparing normal behavior with and without a GRIN lens implant in the target recording area, as well as comparing DBS effects on pathological behavior in animal cohorts with and without a GRIN lens implant, must be completed in future studies. It is important to note that changing the route of access for the GRIN lens intended to image the same target could alter pathological behavior and DBS therapeutic efficacy; therefore, any change to the GRIN lens implantation route would need to be assessed with additional cohorts.